A Deep Learning Based Data Recovery Approach for Missing and Erroneous Data of IoT Nodes

Area of Science:

• Computer Science
• Data Science
• Network Engineering

Background:

• Internet of Things (IoT) nodes generate vast time-series data for automated monitoring.
• Data inconsistency arises from factors like device malfunction, communication instability, and environmental issues.
• Effective data recovery is crucial for maintaining the integrity of IoT data.

Purpose of the Study:

• To propose a novel missing data recovery approach for IoT networks.
• To leverage spatial-temporal (ST) correlations among IoT nodes for data imputation.
• To enhance the reliability of time-series data captured by IoT devices.

Main Methods:

• A two-phase approach: Clustering (CL) and Data Recovery (DR).
• CL phase: Nodes are clustered based on spatial and temporal relationships, identifying common neighbors.
• DR phase: Missing data is recovered using the ST-hierarchical long short-term memory (ST-HLSTM) algorithm, utilizing neighbor node information.

Main Results:

• The proposed ST-HLSTM algorithm demonstrated approximately 98.5% reliability in recovering missing data.
• Validation performed on real-world IoT data from a hydraulic test rig.
• The deep neural network architecture effectively utilizes spatial-temporal features for accurate recovery.

Conclusions:

• The developed ST-HLSTM approach significantly reduces data inconsistency in IoT networks.
• The method offers a reliable solution for missing data imputation in time-series IoT data.
• Spatial-temporal correlation and deep learning are effective in enhancing IoT data quality.